Natural Language to Structured Query

Abstract

Intelligent interface, to enhance e cient interactions between user and databases, is the need of many commercial applications. Oftentimes, users are not familiar with how to frame a structured query as they may not be aware of structure of the database and it is also not expected that the users are required to learn SQL or other query languages to access the database. Hence to simplify task of accessing the database, text-to-SQL models attempt to translate a user's natural language question to corresponding SQL query. Converting natural language to SQL, the model needs to have the ability to create an accurate mapping between the natural language keywords to SQL keywords along with their corresponding tables and columns. Recently, lots of generative text-to-SQL models have been developed. Some of them are using greedy search in their decoder. Hence we choose one of such model[1] and implemented beam search on that. Apart from this we tried explore a discriminative approach for text-to-SQL generation task. A discriminative re-ranker has been proposed on the top of a generative text-to-SQL model for improvement of the accuracy by extracting the best SQL query from a set of beam search predicted candidates. We proposed a schema agnostic discriminative re-ranker built using XLNet ne-tuned classi er for calculating similarity score between natural language and predicted SQL. We used that score to re-rank the beam candidates in a perfect order.